Related Experiment Video
Updated: Jun 25, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
2.7K
Deep learning model for differentiating nasal cavity masses based on nasal endoscopy images
Junhu Tai1, Munsoo Han1,2, Bo Yoon Choi1
1Department of Otorhinolaryngology-Head & Neck Surgery, College of Medicine, Korea University, Seoul, Republic of Korea.
BMC Medical Informatics and Decision Making
|May 29, 2024
Summary
A new deep learning algorithm aids in distinguishing nasal polyps from inverted papillomas using endoscopic images. This computer-aided diagnosis system shows high reliability in identifying lesion locations, improving pre-pathology diagnosis.
Area of Science:
- Otorhinolaryngology
- Medical Imaging
- Artificial Intelligence
Background:
- Nasal polyps and inverted papillomas present diagnostic challenges due to similar endoscopic appearances.
- Accurate differentiation is crucial for appropriate clinical management.
- Current endoscopic examination limitations necessitate advanced diagnostic tools.
Purpose of the Study:
- To develop and evaluate a deep learning algorithm for computer-aided diagnosis of nasal endoscopic images.
- To differentiate between nasal polyps and inverted papillomas.
- To enhance pre-pathologic diagnostic accuracy for nasal masses.
Main Methods:
- Deep learning model trained using curriculum learning on nasal endoscopic images (patches and full-sized).
- Evaluation of a computer-aided diagnosis system for classification of nasal polyps, inverted papilloma, and normal tissue.
- Performance analysis using five-fold cross-validation.
Main Results:
- The model achieved high performance metrics, including an Area Under the Curve (AUC) of 0.97 for normal tissue classification.
- For nasal polyps, the best performance metrics included an AUC of 0.89.
- For inverted papilloma, the best performance metrics included an AUC of 0.83.
Conclusions:
- The developed convolutional neural network demonstrates high reliability in localizing lesions within nasal endoscopic images.
- Gradient-weighted class activation mapping results align with otolaryngologist assessments.
- The deep learning algorithm shows potential for improving clinical diagnosis prior to histopathological confirmation.
